1
0
Fork 0
mirror of https://github.com/ruby/ruby.git synced 2022-11-09 12:17:21 -05:00
ruby--ruby/lib/rexml/parsers/baseparser.rb
ser 7255981a4d r1002 | ser | 2004-06-07 07:45:53 -0400 (Mon, 07 Jun 2004) | 2 lines
* Workin' in the coal mine, goin' down, down, down...
r1003 | ser | 2004-06-08 22:24:08 -0400 (Tue, 08 Jun 2004) | 7 lines
* Entirely rewrote the validation code; the finite state machine, while cool,
  didn't survive the encounter with Interleave.  It was getting sort of hacky,
  too.  The new mechanism is less elegant, but is basically still a FSM, and is
  more flexible without having to add hacks to extend it.  Large chunks of the
  FSM may be reusable in other validation mechanisms.
* Added interleave support
r1004 | ser | 2004-06-09 07:24:17 -0400 (Wed, 09 Jun 2004) | 2 lines
* Added suppert for mixed
r1005 | ser | 2004-06-09 08:01:33 -0400 (Wed, 09 Jun 2004) | 3 lines
* Added Kou's patch to normalize attribute values passed through the SAX2 and
  Stream parsers.
r1006 | ser | 2004-06-09 08:12:35 -0400 (Wed, 09 Jun 2004) | 2 lines
* Applied Kou's preceding-sibling patch, which fixes the order of the axe results
r1009 | ser | 2004-06-20 11:02:55 -0400 (Sun, 20 Jun 2004) | 8 lines
* Redesigned and rewrote the RelaxNG code.  It isn't elegant, but it works.
  Particular problems encountered were interleave and ref.  Interleave means I
  can't use a clean FSM design, and ref means the dirty FSM design has to be modified
  during validation.  There's a lot of code that could be cleaned up in here.
  However, I'm pretty sure that this design is reasonably fast and space efficient.
  I'm not entirely convinced that it is correct; more tests are required.
* This version adds support for defines and refs.
r1011 | ser | 2004-06-20 11:20:07 -0400 (Sun, 20 Jun 2004) | 3 lines
* Removed debugging output from unit test
* Moved ">" in Element.inspect
r1014 | ser | 2004-06-20 11:40:30 -0400 (Sun, 20 Jun 2004) | 2 lines
* Minor big in missing includes for validation rules
r1023 | ser | 2004-07-03 08:57:34 -0400 (Sat, 03 Jul 2004) | 2 lines
* Fixed bug #34, typo in xpath_parser.
r1024 | ser | 2004-07-03 10:22:08 -0400 (Sat, 03 Jul 2004) | 9 lines
* Previous fix, (include? -> includes?) was incorrect.
* Added another test for encoding
* Started AnyName support in RelaxNG
* Added Element#Attributes#to_a, so that it does something intelligent.
  This was needed by XPath, for '@*'
* Fixed XPath so that @* works.


git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@6577 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2004-07-04 15:26:07 +00:00

451 lines
15 KiB
Ruby

require 'rexml/parseexception'
require 'rexml/source'
module REXML
module Parsers
# = Using the Pull Parser
# <em>This API is experimental, and subject to change.</em>
# parser = PullParser.new( "<a>text<b att='val'/>txet</a>" )
# while parser.has_next?
# res = parser.next
# puts res[1]['att'] if res.start_tag? and res[0] == 'b'
# end
# See the PullEvent class for information on the content of the results.
# The data is identical to the arguments passed for the various events to
# the StreamListener API.
#
# Notice that:
# parser = PullParser.new( "<a>BAD DOCUMENT" )
# while parser.has_next?
# res = parser.next
# raise res[1] if res.error?
# end
#
# Nat Price gave me some good ideas for the API.
class BaseParser
NCNAME_STR= '[\w:][\-\w\d.]*'
NAME_STR= "(?:#{NCNAME_STR}:)?#{NCNAME_STR}"
NAMECHAR = '[\-\w\d\.:]'
NAME = "([\\w:]#{NAMECHAR}*)"
NMTOKEN = "(?:#{NAMECHAR})+"
NMTOKENS = "#{NMTOKEN}(\\s+#{NMTOKEN})*"
REFERENCE = "(?:&#{NAME};|&#\\d+;|&#x[0-9a-fA-F]+;)"
REFERENCE_RE = /#{REFERENCE}/
DOCTYPE_START = /\A\s*<!DOCTYPE\s/um
DOCTYPE_PATTERN = /\s*<!DOCTYPE\s+(.*?)(\[|>)/um
ATTRIBUTE_PATTERN = /\s*(#{NAME_STR})\s*=\s*(["'])(.*?)\2/um
COMMENT_START = /\A<!--/u
COMMENT_PATTERN = /<!--(.*?)-->/um
CDATA_START = /\A<!\[CDATA\[/u
CDATA_END = /^\s*\]\s*>/um
CDATA_PATTERN = /<!\[CDATA\[(.*?)\]\]>/um
XMLDECL_START = /\A<\?xml\s/u;
XMLDECL_PATTERN = /<\?xml\s+(.*?)\?>*/um
INSTRUCTION_START = /\A<\?/u
INSTRUCTION_PATTERN = /<\?(.*?)(\s+.*?)?\?>/um
TAG_MATCH = /^<((?>#{NAME_STR}))\s*((?>\s+#{NAME_STR}\s*=\s*(["']).*?\3)*)\s*(\/)?>/um
CLOSE_MATCH = /^\s*<\/(#{NAME_STR})\s*>/um
VERSION = /\bversion\s*=\s*["'](.*?)['"]/um
ENCODING = /\bencoding=["'](.*?)['"]/um
STANDALONE = /\bstandalone=["'](.*?)['"]/um
ENTITY_START = /^\s*<!ENTITY/
IDENTITY = /^([!\*\w\-]+)(\s+#{NCNAME_STR})?(\s+["'].*?['"])?(\s+['"].*?["'])?/u
ELEMENTDECL_START = /^\s*<!ELEMENT/um
ELEMENTDECL_PATTERN = /^\s*(<!ELEMENT.*?)>/um
SYSTEMENTITY = /^\s*(%.*?;)\s*$/um
ENUMERATION = "\\(\\s*#{NMTOKEN}(?:\\s*\\|\\s*#{NMTOKEN})*\\s*\\)"
NOTATIONTYPE = "NOTATION\\s+\\(\\s*#{NAME}(?:\\s*\\|\\s*#{NAME})*\\s*\\)"
ENUMERATEDTYPE = "(?:(?:#{NOTATIONTYPE})|(?:#{ENUMERATION}))"
ATTTYPE = "(CDATA|ID|IDREF|IDREFS|ENTITY|ENTITIES|NMTOKEN|NMTOKENS|#{ENUMERATEDTYPE})"
ATTVALUE = "(?:\"((?:[^<&\"]|#{REFERENCE})*)\")|(?:'((?:[^<&']|#{REFERENCE})*)')"
DEFAULTDECL = "(#REQUIRED|#IMPLIED|(?:(#FIXED\\s+)?#{ATTVALUE}))"
ATTDEF = "\\s+#{NAME}\\s+#{ATTTYPE}\\s+#{DEFAULTDECL}"
ATTDEF_RE = /#{ATTDEF}/
ATTLISTDECL_START = /^\s*<!ATTLIST/um
ATTLISTDECL_PATTERN = /^\s*<!ATTLIST\s+#{NAME}(?:#{ATTDEF})*\s*>/um
NOTATIONDECL_START = /^\s*<!NOTATION/um
PUBLIC = /^\s*<!NOTATION\s+(\w[\-\w]*)\s+(PUBLIC)\s+((["']).*?\4)\s*>/um
SYSTEM = /^\s*<!NOTATION\s+(\w[\-\w]*)\s+(SYSTEM)\s+((["']).*?\4)\s*>/um
TEXT_PATTERN = /\A([^<]*)/um
# Entity constants
PUBIDCHAR = "\x20\x0D\x0Aa-zA-Z0-9\\-()+,./:=?;!*@$_%#"
SYSTEMLITERAL = %Q{((?:"[^"]*")|(?:'[^']*'))}
PUBIDLITERAL = %Q{("[#{PUBIDCHAR}']*"|'[#{PUBIDCHAR}]*')}
EXTERNALID = "(?:(?:(SYSTEM)\\s+#{SYSTEMLITERAL})|(?:(PUBLIC)\\s+#{PUBIDLITERAL}\\s+#{SYSTEMLITERAL}))"
NDATADECL = "\\s+NDATA\\s+#{NAME}"
PEREFERENCE = "%#{NAME};"
ENTITYVALUE = %Q{((?:"(?:[^%&"]|#{PEREFERENCE}|#{REFERENCE})*")|(?:'([^%&']|#{PEREFERENCE}|#{REFERENCE})*'))}
PEDEF = "(?:#{ENTITYVALUE}|#{EXTERNALID})"
ENTITYDEF = "(?:#{ENTITYVALUE}|(?:#{EXTERNALID}(#{NDATADECL})?))"
PEDECL = "<!ENTITY\\s+(%)\\s+#{NAME}\\s+#{PEDEF}\\s*>"
GEDECL = "<!ENTITY\\s+#{NAME}\\s+#{ENTITYDEF}\\s*>"
ENTITYDECL = /\s*(?:#{GEDECL})|(?:#{PEDECL})/um
EREFERENCE = /&(?!#{NAME};)/
DEFAULT_ENTITIES = {
'gt' => [/&gt;/, '&gt;', '>', />/],
'lt' => [/&lt;/, '&lt;', '<', /</],
'quot' => [/&quot;/, '&quot;', '"', /"/],
"apos" => [/&apos;/, "&apos;", "'", /'/]
}
def initialize( source )
self.stream = source
end
def add_listener( listener )
if !defined?(@listeners) or !@listeners
@listeners = []
instance_eval <<-EOL
alias :_old_pull :pull
def pull
event = _old_pull
@listeners.each do |listener|
listener.receive event
end
event
end
EOL
end
@listeners << listener
end
attr_reader :source
def stream=( source )
if source.kind_of? String
@source = Source.new(source)
elsif source.kind_of? IO
@source = IOSource.new(source)
elsif source.kind_of? Source
@source = source
elsif defined? StringIO and source.kind_of? StringIO
@source = IOSource.new(source)
else
raise "#{source.class} is not a valid input stream. It must be \n"+
"either a String, IO, StringIO or Source."
end
@closed = nil
@document_status = nil
@tags = []
@stack = []
@entities = []
end
# Returns true if there are no more events
def empty?
#puts "@source.empty? = #{@source.empty?}"
#puts "@stack.empty? = #{@stack.empty?}"
return (@source.empty? and @stack.empty?)
end
# Returns true if there are more events. Synonymous with !empty?
def has_next?
return !(@source.empty? and @stack.empty?)
end
# Push an event back on the head of the stream. This method
# has (theoretically) infinite depth.
def unshift token
@stack.unshift(token)
end
# Peek at the +depth+ event in the stack. The first element on the stack
# is at depth 0. If +depth+ is -1, will parse to the end of the input
# stream and return the last event, which is always :end_document.
# Be aware that this causes the stream to be parsed up to the +depth+
# event, so you can effectively pre-parse the entire document (pull the
# entire thing into memory) using this method.
def peek depth=0
raise %Q[Illegal argument "#{depth}"] if depth < -1
temp = []
if depth == -1
temp.push(pull()) until empty?
else
while @stack.size+temp.size < depth+1
temp.push(pull())
end
end
@stack += temp if temp.size > 0
@stack[depth]
end
# Returns the next event. This is a +PullEvent+ object.
def pull
if @closed
x, @closed = @closed, nil
return [ :end_element, x ]
end
return [ :end_document ] if empty?
return @stack.shift if @stack.size > 0
@source.read if @source.buffer.size<2
if @document_status == nil
@source.consume( /^\s*/um )
word = @source.match( /(<[^>]*)>/um )
word = word[1] unless word.nil?
case word
when COMMENT_START
return [ :comment, @source.match( COMMENT_PATTERN, true )[1] ]
when XMLDECL_START
results = @source.match( XMLDECL_PATTERN, true )[1]
version = VERSION.match( results )
version = version[1] unless version.nil?
encoding = ENCODING.match(results)
encoding = encoding[1] unless encoding.nil?
@source.encoding = encoding
standalone = STANDALONE.match(results)
standalone = standalone[1] unless standalone.nil?
return [ :xmldecl, version, encoding, standalone]
when INSTRUCTION_START
return [ :processing_instruction, *@source.match(INSTRUCTION_PATTERN, true)[1,2] ]
when DOCTYPE_START
md = @source.match( DOCTYPE_PATTERN, true )
identity = md[1]
close = md[2]
identity =~ IDENTITY
name = $1
raise REXML::ParseException("DOCTYPE is missing a name") if name.nil?
pub_sys = $2.nil? ? nil : $2.strip
long_name = $3.nil? ? nil : $3.strip
uri = $4.nil? ? nil : $4.strip
args = [ :start_doctype, name, pub_sys, long_name, uri ]
if close == ">"
@document_status = :after_doctype
@source.read if @source.buffer.size<2
md = @source.match(/^\s*/um, true)
@stack << [ :end_doctype ]
else
@document_status = :in_doctype
end
return args
else
@document_status = :after_doctype
@source.read if @source.buffer.size<2
md = @source.match(/\s*/um, true)
end
end
if @document_status == :in_doctype
md = @source.match(/\s*(.*?>)/um)
case md[1]
when SYSTEMENTITY
match = @source.match( SYSTEMENTITY, true )[1]
return [ :externalentity, match ]
when ELEMENTDECL_START
return [ :elementdecl, @source.match( ELEMENTDECL_PATTERN, true )[1] ]
when ENTITY_START
match = @source.match( ENTITYDECL, true ).to_a.compact
match[0] = :entitydecl
ref = false
if match[1] == '%'
ref = true
match.delete_at 1
end
# Now we have to sort out what kind of entity reference this is
if match[2] == 'SYSTEM'
# External reference
match[3] = match[3][1..-2] # PUBID
match.delete_at(4) if match.size > 4 # Chop out NDATA decl
# match is [ :entity, name, SYSTEM, pubid(, ndata)? ]
elsif match[2] == 'PUBLIC'
# External reference
match[3] = match[3][1..-2] # PUBID
match[4] = match[4][1..-2] # HREF
# match is [ :entity, name, PUBLIC, pubid, href ]
else
match[2] = match[2][1..-2]
match.pop if match.size == 4
# match is [ :entity, name, value ]
end
match << '%' if ref
return match
when ATTLISTDECL_START
md = @source.match( ATTLISTDECL_PATTERN, true )
raise REXML::ParseException.new( "Bad ATTLIST declaration!", @source ) if md.nil?
element = md[1]
contents = md[0]
pairs = {}
values = md[0].scan( ATTDEF_RE )
values.each do |attdef|
unless attdef[3] == "#IMPLIED"
attdef.compact!
val = attdef[3]
val = attdef[4] if val == "#FIXED "
pairs[attdef[0]] = val
end
end
return [ :attlistdecl, element, pairs, contents ]
when NOTATIONDECL_START
md = nil
if @source.match( PUBLIC )
md = @source.match( PUBLIC, true )
elsif @source.match( SYSTEM )
md = @source.match( SYSTEM, true )
else
raise REXML::ParseException.new( "error parsing notation: no matching pattern", @source )
end
return [ :notationdecl, md[1], md[2], md[3] ]
when CDATA_END
@document_status = :after_doctype
@source.match( CDATA_END, true )
return [ :end_doctype ]
end
end
begin
if @source.buffer[0] == ?<
if @source.buffer[1] == ?/
last_tag = @tags.pop
#md = @source.match_to_consume( '>', CLOSE_MATCH)
md = @source.match( CLOSE_MATCH, true )
raise REXML::ParseException.new( "Missing end tag for "+
"'#{last_tag}' (got \"#{md[1]}\")",
@source) unless last_tag == md[1]
return [ :end_element, last_tag ]
elsif @source.buffer[1] == ?!
md = @source.match(/\A(\s*[^>]*>)/um)
#puts "SOURCE BUFFER = #{source.buffer}, #{source.buffer.size}"
raise REXML::ParseException.new("Malformed node", @source) unless md
if md[0][2] == ?-
md = @source.match( COMMENT_PATTERN, true )
return [ :comment, md[1] ] if md
else
md = @source.match( CDATA_PATTERN, true )
return [ :cdata, md[1] ] if md
end
raise REXML::ParseException.new( "Declarations can only occur "+
"in the doctype declaration.", @source)
elsif @source.buffer[1] == ??
md = @source.match( INSTRUCTION_PATTERN, true )
return [ :processing_instruction, md[1], md[2] ] if md
raise REXML::ParseException.new( "Bad instruction declaration",
@source)
else
# Get the next tag
md = @source.match(TAG_MATCH, true)
raise REXML::ParseException.new("malformed XML: missing tag start", @source) unless md
attrs = []
if md[2].size > 0
attrs = md[2].scan( ATTRIBUTE_PATTERN )
raise REXML::ParseException.new( "error parsing attributes: [#{attrs.join ', '}], excess = \"#$'\"", @source) if $' and $'.strip.size > 0
end
if md[4]
@closed = md[1]
else
@tags.push( md[1] )
end
attributes = {}
attrs.each { |a,b,c| attributes[a] = c }
return [ :start_element, md[1], attributes ]
end
else
md = @source.match( TEXT_PATTERN, true )
if md[0].length == 0
#puts "EMPTY = #{empty?}"
#puts "BUFFER = \"#{@source.buffer}\""
@source.match( /(\s+)/, true )
end
#return [ :text, "" ] if md[0].length == 0
# unnormalized = Text::unnormalize( md[1], self )
# return PullEvent.new( :text, md[1], unnormalized )
return [ :text, md[1] ]
end
rescue REXML::ParseException
raise
rescue Exception, NameError => error
raise REXML::ParseException.new( "Exception parsing",
@source, self, (error ? error : $!) )
end
return [ :dummy ]
end
def entity( reference, entities )
value = nil
value = entities[ reference ] if entities
if not value
value = DEFAULT_ENTITIES[ reference ]
value = value[2] if value
end
unnormalize( value, entities ) if value
end
# Escapes all possible entities
def normalize( input, entities=nil, entity_filter=nil )
copy = input.clone
# Doing it like this rather than in a loop improves the speed
copy.gsub!( EREFERENCE, '&amp;' )
entities.each do |key, value|
copy.gsub!( value, "&#{key};" ) unless entity_filter and
entity_filter.include?(entity)
end if entities
copy.gsub!( EREFERENCE, '&amp;' )
DEFAULT_ENTITIES.each do |key, value|
copy.gsub!( value[3], value[1] )
end
copy
end
# Unescapes all possible entities
def unnormalize( string, entities=nil, filter=nil )
rv = string.clone
rv.gsub!( /\r\n?/, "\n" )
matches = rv.scan( REFERENCE_RE )
return rv if matches.size == 0
rv.gsub!( /&#0*((?:\d+)|(?:x[a-fA-F0-9]+));/ ) {|m|
m=$1
m = "0#{m}" if m[0] == ?x
[Integer(m)].pack('U*')
}
matches.collect!{|x|x[0]}.compact!
if matches.size > 0
matches.each do |entity_reference|
unless filter and filter.include?(entity_reference)
entity_value = entity( entity_reference, entities )
if entity_value
re = /&#{entity_reference};/
rv.gsub!( re, entity_value )
end
end
end
matches.each do |entity_reference|
unless filter and filter.include?(entity_reference)
er = DEFAULT_ENTITIES[entity_reference]
rv.gsub!( er[0], er[2] ) if er
end
end
rv.gsub!( /&amp;/, '&' )
end
rv
end
end
end
end
=begin
case event[0]
when :start_element
when :text
when :end_element
when :processing_instruction
when :cdata
when :comment
when :xmldecl
when :start_doctype
when :end_doctype
when :externalentity
when :elementdecl
when :entity
when :attlistdecl
when :notationdecl
when :end_doctype
end
=end